Crew Resource Management for Automated Teammates (CRM-A) Robert J. Shively1*, Joel Lachter1, Robert Koteskey2 and Summer L. Brandt2 1 NASA Ames Research Center, Moffett Field CA 94040, USA 2San Jose State University, Moffett Field CA 94040, USA {robert.j.shively, joel.lachter, robert.w.koteskey, summer.l.brandt}@nasa.gov Abstract. Crew Resource Management (CRM) is the application of human fac- tors knowledge and skills to ensure that teams make effective use of all re- sources. This includes ensuring that pilots bring in opinions of other teammates and utilize their unique capabilities. CRM was originally developed 40 years ago in response to a number of airline accidents in which the crew was found to be at fault. The goal was to improve teamwork among airline cockpit crews. The notion of “team” was later expanded to include cabin crew and ground re- sources. CRM has also been adopted by other industries, most notably medi- cine. Automation research now finds itself faced with similar issues to those faced by aviation 40 years ago: how to create a more robust system by making full use of both the automation and its human operators. With advances in ma- chine intelligence, processing speed and cheap and plentiful memory, automa- tion has advanced to the point that it can and should be treated as a teammate to fully take advantage of its capabilities and contributions to the system. This area of research is known as Human-Autonomy Teaming (HAT). Research on HAT has identified reusable patterns that can be applied in a wide range of applica- tions. These patterns include features such as bi-directional communication and working agreements. This paper will explore the synergies between CRM and HAT. We believe that HAT research has much to learn from CRM and that there are benefits to expanding CRM to cover automation. Keywords: Crew Resource Management (CRM), Human-Autonomy Teaming (HAT), Automation. 1 Why Human-Autonomy Teaming (HAT)? For centuries, automation has been making humanity more productive. Historically this automation has chiefly replaced the brawn of human labor, leaving people as the brains. Recently, however, machines have been used by humans for increasingly cog- nitive tasks. Computers can play chess better than humans. They can search the web to answer your questions. They can drive cars. But automation is not perfect. It is brittle, breaking, often catastrophically, when taken out of the “comfort zone” it was designed for. It has crashed stock exchanges [1] and cars [2]. Self-driving cars still need to “phone home” when encountering a person directing traffic [3]. Aviation has not been immune to this increase in automation, from relatively simple autopilots to sophisticated flight management systems. These innovations have not always been smooth [4]; however, the development of automation as tools to improve flight safety and efficiency has continued. As the nature of automation changes, so should the role of the human when interacting with automation. Automation is moving from the realm of simple “tools” into intelligent cognitive agents that can function as teammates, similar to human teammates. Innovations in artificial intelligence as well as increases in the speed and memory of the underlying hardware have spurred this shift. Cognitive agents can now suggest courses of action, monitor the operator actions and physiology, and monitor the environment. With all of these new capabilities, work has begun to investigate how the relationship of the human and the automation can be shifted toward that of teammates to best take advantage of this phenomenon. This idea of Human- Autonomy Teaming (HAT) goes beyond simply giving a person a computer. The idea is that humans and automation should work together the way (well-functioning) human teams do, bouncing ideas off of one another, backing each other up when they sense potential problems, and keeping each other informed of what they are doing. HAT is currently recognized as a promising solution to the problems of human operators managing increasingly complex work systems. A human-autonomy team has been defined as “the dynamic, interdependent coupling between one or more human operators and one or more automated systems requiring collaboration and coordination to achieve successful task completion;” [5] a definition that has been picked up by others [6, 7]. As such, it is being developed and pursued in many operational areas such as robotics [8], commercial aviation [9], and UAS operations [10]. Aviation, with its very systematic approach to safety, may prove to be both a source of inspiration for developing better human-autonomy teaming, and an industry where its benefits might be most profitably exploited. While computers have eliminated the need for the navigator and (together with an increase in reliability) the flight engineer, regulatory barriers have slowed the ascent of automation in the cockpit. Years of cultivating teamwork on the flight deck have contributed to aviation’s superb safety record. Here we explore potential synergies between the lessons the airline industry has learned from years of studying teamwork and this newer field of human-autonomy teaming. Can we make human-autonomy teaming more effective by looking at the aviation model? Can we introduce advanced automation to the flight deck more safely if we introduce it as a team member? In the cockpit, where the importance of teaming has long been understood, the skills necessary for good teaming and the training procedures for teaching those skills have been codified under the umbrella of Crew Resource Management (CRM). Here we explore the synergies between HAT and CRM that may allow these new computerized teammates to be incorporated into CRM and may improve HAT concepts by incorporating the lessons learned from the development of CRM. 2 What is Crew Resource Management (CRM)? 2.1 History of CRM in Human Teams What can we learn from CRM as it has been developed for human teams? Although the discipline has expanded to other high-risk high-reliability industries, the multi- crew airline flight deck is where CRM has its roots. Risk management has always been the core task for an airline pilot. In the early days of transport aviation, risks due to mechanical failure were more prominent and aircraft were less reliable. The majority of threats were clear, and external to the human who was tasked with operating the machine. Early airline captains were solo performers whose technical skills were sharpened by absolute necessity. Their selection, their environment, and their culture reinforced strong, independent personalities and isolated, individual decision making. As large aircraft became more complex, a single pilot could no longer operate the aircraft. The crew compliment grew but was comprised of people who were still focused on individual tasks. As the job quickly became a more team-oriented endeavor, flight deck culture and nature of the individual pilot had not necessarily changed to reflect this shift. By the late 1960's and 1970's transport accidents due to mechanical failure had drastically decreased. Advances in aviation technology like jet engines, modern avionics, and increased automation, so lowered the accident rate that the majority of new occurrences were now found to be crew related. Thus, further safety improvement could most easily be found from within the human team [11]. Landmark accidents like UAL 173, caused by fuel exhaustion, and EAL 401, brought down by the distraction from a faulty landing gear indicator, highlighted the need for training on team leadership and decision making for multi-crew aircraft [12]. The rugged, isolated individual was perhaps not the ideal model for what was now clearly a team activity. A breakthrough study, conducted by Ruffel Smith [13] suggested a correlation be- tween the leadership and communication style of the captain with overall crew per- formance. Further research reinforced this hypothesis [14, 15], and during the 1980's, industry and government came together to form what was at the time, a new discipline of aviation team training called Cockpit Resource Management, or CRM. As it evolved and became recognized as applicable to the larger aviation community, it was renamed Crew Resource Management. CRM has been defined as “using all available resources-information, equipment, and people-to achieve safe and efficient flight operations” [16, pg. 20]. In the 1990's, CRM training programs were introduced at major airlines and in military aviation units. Many of the concepts originally trained were lifted from business school management training templates and were not particularly well received by pilots. The topics in these classroom seminars typically included assessments of personality and leadership style. While these assessments might have been good predictors of performance, they did not necessarily allow for actual behavioral change. The goal of the training was an attitudinal shift, but this was hard to measure and equally difficult to reliably move to the flight deck. Eventually, over the course of two decades, emphasis has shifted to identifying observable flight deck team behaviors (good and bad) that could be trained and evaluated in actual line operations. Task analyses were conducted at major carriers and research organizations. Both technical and CRM skills were identified and then used to create training in high fidelity, full flight, line-oriented simulations. These Line Oriented Flight Training (LOFT) events are now a primary means of introducing, reinforcing, and evaluating CRM skills [17]. CRM training today still includes an indoctrination seminar for new hires, but now CRM skills are embedded in all aspects of multi-crew transport training. They are trained and evaluated alongside technical skills at every training opportunity. Current CRM doctrine uses the concept of Threat and Error Management (TEM). This paradigm seeks to acknowledge the universal existence of human error and outside threats. TEM seeks to engage the team in actively searching for those threats and inevitable errors, then through use of CRM skills, to avoid, minimize or correct them.